xref: /aosp_15_r20/external/llvm-libc/src/__support/FPUtil/ManipulationFunctions.h (revision 71db0c75aadcf003ffe3238005f61d7618a3fead)
1 //===-- Floating-point manipulation functions -------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #ifndef LLVM_LIBC_SRC___SUPPORT_FPUTIL_MANIPULATIONFUNCTIONS_H
10 #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_MANIPULATIONFUNCTIONS_H
11 
12 #include "FPBits.h"
13 #include "NearestIntegerOperations.h"
14 #include "NormalFloat.h"
15 #include "cast.h"
16 #include "dyadic_float.h"
17 #include "rounding_mode.h"
18 
19 #include "hdr/math_macros.h"
20 #include "src/__support/CPP/bit.h"
21 #include "src/__support/CPP/limits.h" // INT_MAX, INT_MIN
22 #include "src/__support/CPP/type_traits.h"
23 #include "src/__support/FPUtil/FEnvImpl.h"
24 #include "src/__support/macros/attributes.h"
25 #include "src/__support/macros/config.h"
26 #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY
27 
28 namespace LIBC_NAMESPACE_DECL {
29 namespace fputil {
30 
31 template <typename T, cpp::enable_if_t<cpp::is_floating_point_v<T>, int> = 0>
frexp(T x,int & exp)32 LIBC_INLINE T frexp(T x, int &exp) {
33   FPBits<T> bits(x);
34   if (bits.is_inf_or_nan()) {
35 #ifdef LIBC_FREXP_INF_NAN_EXPONENT
36     // The value written back to the second parameter when calling
37     // frexp/frexpf/frexpl` with `+/-Inf`/`NaN` is unspecified in the standard.
38     // Set the exp value for Inf/NaN inputs explicitly to
39     // LIBC_FREXP_INF_NAN_EXPONENT if it is defined.
40     exp = LIBC_FREXP_INF_NAN_EXPONENT;
41 #endif // LIBC_FREXP_INF_NAN_EXPONENT
42     return x;
43   }
44   if (bits.is_zero()) {
45     exp = 0;
46     return x;
47   }
48 
49   NormalFloat<T> normal(bits);
50   exp = normal.exponent + 1;
51   normal.exponent = -1;
52   return normal;
53 }
54 
55 template <typename T, cpp::enable_if_t<cpp::is_floating_point_v<T>, int> = 0>
modf(T x,T & iptr)56 LIBC_INLINE T modf(T x, T &iptr) {
57   FPBits<T> bits(x);
58   if (bits.is_zero() || bits.is_nan()) {
59     iptr = x;
60     return x;
61   } else if (bits.is_inf()) {
62     iptr = x;
63     return FPBits<T>::zero(bits.sign()).get_val();
64   } else {
65     iptr = trunc(x);
66     if (x == iptr) {
67       // If x is already an integer value, then return zero with the right
68       // sign.
69       return FPBits<T>::zero(bits.sign()).get_val();
70     } else {
71       return x - iptr;
72     }
73   }
74 }
75 
76 template <typename T, cpp::enable_if_t<cpp::is_floating_point_v<T>, int> = 0>
copysign(T x,T y)77 LIBC_INLINE T copysign(T x, T y) {
78   FPBits<T> xbits(x);
79   xbits.set_sign(FPBits<T>(y).sign());
80   return xbits.get_val();
81 }
82 
83 template <typename T> struct IntLogbConstants;
84 
85 template <> struct IntLogbConstants<int> {
86   LIBC_INLINE_VAR static constexpr int FP_LOGB0 = FP_ILOGB0;
87   LIBC_INLINE_VAR static constexpr int FP_LOGBNAN = FP_ILOGBNAN;
88   LIBC_INLINE_VAR static constexpr int T_MAX = INT_MAX;
89   LIBC_INLINE_VAR static constexpr int T_MIN = INT_MIN;
90 };
91 
92 template <> struct IntLogbConstants<long> {
93   LIBC_INLINE_VAR static constexpr long FP_LOGB0 = FP_ILOGB0;
94   LIBC_INLINE_VAR static constexpr long FP_LOGBNAN = FP_ILOGBNAN;
95   LIBC_INLINE_VAR static constexpr long T_MAX = LONG_MAX;
96   LIBC_INLINE_VAR static constexpr long T_MIN = LONG_MIN;
97 };
98 
99 template <typename T, typename U>
100 LIBC_INLINE constexpr cpp::enable_if_t<cpp::is_floating_point_v<U>, T>
101 intlogb(U x) {
102   FPBits<U> bits(x);
103   if (LIBC_UNLIKELY(bits.is_zero() || bits.is_inf_or_nan())) {
104     set_errno_if_required(EDOM);
105     raise_except_if_required(FE_INVALID);
106 
107     if (bits.is_zero())
108       return IntLogbConstants<T>::FP_LOGB0;
109     if (bits.is_nan())
110       return IntLogbConstants<T>::FP_LOGBNAN;
111     // bits is inf.
112     return IntLogbConstants<T>::T_MAX;
113   }
114 
115   DyadicFloat<FPBits<U>::STORAGE_LEN> normal(bits.get_val());
116   int exponent = normal.get_unbiased_exponent();
117   // The C standard does not specify the return value when an exponent is
118   // out of int range. However, XSI conformance required that INT_MAX or
119   // INT_MIN are returned.
120   // NOTE: It is highly unlikely that exponent will be out of int range as
121   // the exponent is only 15 bits wide even for the 128-bit floating point
122   // format.
123   if (LIBC_UNLIKELY(exponent > IntLogbConstants<T>::T_MAX ||
124                     exponent < IntLogbConstants<T>::T_MIN)) {
125     set_errno_if_required(ERANGE);
126     raise_except_if_required(FE_INVALID);
127     return exponent > 0 ? IntLogbConstants<T>::T_MAX
128                         : IntLogbConstants<T>::T_MIN;
129   }
130 
131   return static_cast<T>(exponent);
132 }
133 
134 template <typename T, cpp::enable_if_t<cpp::is_floating_point_v<T>, int> = 0>
135 LIBC_INLINE constexpr T logb(T x) {
136   FPBits<T> bits(x);
137   if (LIBC_UNLIKELY(bits.is_zero() || bits.is_inf_or_nan())) {
138     if (bits.is_nan())
139       return x;
140 
141     raise_except_if_required(FE_DIVBYZERO);
142 
143     if (bits.is_zero()) {
144       set_errno_if_required(ERANGE);
145       return FPBits<T>::inf(Sign::NEG).get_val();
146     }
147     // bits is inf.
148     return FPBits<T>::inf().get_val();
149   }
150 
151   DyadicFloat<FPBits<T>::STORAGE_LEN> normal(bits.get_val());
152   return static_cast<T>(normal.get_unbiased_exponent());
153 }
154 
155 template <typename T, typename U>
156 LIBC_INLINE constexpr cpp::enable_if_t<
157     cpp::is_floating_point_v<T> && cpp::is_integral_v<U>, T>
158 ldexp(T x, U exp) {
159   FPBits<T> bits(x);
160   if (LIBC_UNLIKELY((exp == 0) || bits.is_zero() || bits.is_inf_or_nan()))
161     return x;
162 
163   // NormalFloat uses int32_t to store the true exponent value. We should ensure
164   // that adding |exp| to it does not lead to integer rollover. But, if |exp|
165   // value is larger the exponent range for type T, then we can return infinity
166   // early. Because the result of the ldexp operation can be a subnormal number,
167   // we need to accommodate the (mantissaWidth + 1) worth of shift in
168   // calculating the limit.
169   constexpr int EXP_LIMIT =
170       FPBits<T>::MAX_BIASED_EXPONENT + FPBits<T>::FRACTION_LEN + 1;
171   // Make sure that we can safely cast exp to int when not returning early.
172   static_assert(EXP_LIMIT <= INT_MAX && -EXP_LIMIT >= INT_MIN);
173   if (LIBC_UNLIKELY(exp > EXP_LIMIT)) {
174     int rounding_mode = quick_get_round();
175     Sign sign = bits.sign();
176 
177     if ((sign == Sign::POS && rounding_mode == FE_DOWNWARD) ||
178         (sign == Sign::NEG && rounding_mode == FE_UPWARD) ||
179         (rounding_mode == FE_TOWARDZERO))
180       return FPBits<T>::max_normal(sign).get_val();
181 
182     set_errno_if_required(ERANGE);
183     raise_except_if_required(FE_OVERFLOW);
184     return FPBits<T>::inf(sign).get_val();
185   }
186 
187   // Similarly on the negative side we return zero early if |exp| is too small.
188   if (LIBC_UNLIKELY(exp < -EXP_LIMIT)) {
189     int rounding_mode = quick_get_round();
190     Sign sign = bits.sign();
191 
192     if ((sign == Sign::POS && rounding_mode == FE_UPWARD) ||
193         (sign == Sign::NEG && rounding_mode == FE_DOWNWARD))
194       return FPBits<T>::min_subnormal(sign).get_val();
195 
196     set_errno_if_required(ERANGE);
197     raise_except_if_required(FE_UNDERFLOW);
198     return FPBits<T>::zero(sign).get_val();
199   }
200 
201   // For all other values, NormalFloat to T conversion handles it the right way.
202   DyadicFloat<FPBits<T>::STORAGE_LEN> normal(bits.get_val());
203   normal.exponent += static_cast<int>(exp);
204   // TODO: Add tests for exceptions.
205   return normal.template as<T, /*ShouldRaiseExceptions=*/true>();
206 }
207 
208 template <typename T, typename U,
209           cpp::enable_if_t<cpp::is_floating_point_v<T> &&
210                                cpp::is_floating_point_v<U> &&
211                                (sizeof(T) <= sizeof(U)),
212                            int> = 0>
213 LIBC_INLINE T nextafter(T from, U to) {
214   FPBits<T> from_bits(from);
215   if (from_bits.is_nan())
216     return from;
217 
218   FPBits<U> to_bits(to);
219   if (to_bits.is_nan())
220     return cast<T>(to);
221 
222   // NOTE: This would work only if `U` has a greater or equal precision than
223   // `T`. Otherwise `from` could loose its precision and the following statement
224   // could incorrectly evaluate to `true`.
225   if (cast<U>(from) == to)
226     return cast<T>(to);
227 
228   using StorageType = typename FPBits<T>::StorageType;
229   if (from != T(0)) {
230     if ((cast<U>(from) < to) == (from > T(0))) {
231       from_bits = FPBits<T>(StorageType(from_bits.uintval() + 1));
232     } else {
233       from_bits = FPBits<T>(StorageType(from_bits.uintval() - 1));
234     }
235   } else {
236     from_bits = FPBits<T>::min_subnormal(to_bits.sign());
237   }
238 
239   if (from_bits.is_subnormal())
240     raise_except_if_required(FE_UNDERFLOW | FE_INEXACT);
241   else if (from_bits.is_inf())
242     raise_except_if_required(FE_OVERFLOW | FE_INEXACT);
243 
244   return from_bits.get_val();
245 }
246 
247 template <bool IsDown, typename T,
248           cpp::enable_if_t<cpp::is_floating_point_v<T>, int> = 0>
249 LIBC_INLINE constexpr T nextupdown(T x) {
250   constexpr Sign sign = IsDown ? Sign::NEG : Sign::POS;
251 
252   FPBits<T> xbits(x);
253   if (xbits.is_nan() || xbits == FPBits<T>::max_normal(sign) ||
254       xbits == FPBits<T>::inf(sign))
255     return x;
256 
257   using StorageType = typename FPBits<T>::StorageType;
258   if (x != T(0)) {
259     if (xbits.sign() == sign) {
260       xbits = FPBits<T>(StorageType(xbits.uintval() + 1));
261     } else {
262       xbits = FPBits<T>(StorageType(xbits.uintval() - 1));
263     }
264   } else {
265     xbits = FPBits<T>::min_subnormal(sign);
266   }
267 
268   return xbits.get_val();
269 }
270 
271 } // namespace fputil
272 } // namespace LIBC_NAMESPACE_DECL
273 
274 #ifdef LIBC_TYPES_LONG_DOUBLE_IS_X86_FLOAT80
275 #include "x86_64/NextAfterLongDouble.h"
276 #include "x86_64/NextUpDownLongDouble.h"
277 #endif // LIBC_TYPES_LONG_DOUBLE_IS_X86_FLOAT80
278 
279 #endif // LLVM_LIBC_SRC___SUPPORT_FPUTIL_MANIPULATIONFUNCTIONS_H
280